Prompt Sensitivity

Prompt sensitivity describes how much a model’s output can change based on seemingly minor variations in how you phrase a request. Asking “Can you summarize this?” versus “Please summarize this” might produce different responses; adding or removing a comma can shift tone; putting instructions at the beginning versus the end of a prompt can change whether they’re followed. High prompt sensitivity is a sign of fragile behavior — outputs that are unreliable in production, where user inputs are naturally variable. For behavior architects, sensitivity testing is an important part of prompt validation: before relying on a prompt in a product, you should test it with a range of realistic phrasings to ensure it degrades gracefully rather than failing unpredictably.